Something like this might serve:

>perl -wMstrict -le
"print q{output:};
 for my $url (@ARGV) {
   (my $filename = $url) =~
     s{ .* / (\d+) / (_ \d+ \. pdf) \z } {$1\L$2}xmsi;

   print qq{url: $url -> filename: $filename};
   }
"
/pdf/100004/_20090326.pdf
/pdf/100006/_20090326.pdf
/foo/bar/pdf/123/_456.PdF
output:
url: /pdf/100004/_20090326.pdf -> filename: 100004_20090326.pdf
url: /pdf/100006/_20090326.pdf -> filename: 100006_20090326.pdf
url: /foo/bar/pdf/123/_456.PdF -> filename: 123_456.pdf
[download]

Or a slightly simpler version using split (output still lowercased):

>perl -wMstrict -le
"print q{output:};
 for my $url (@ARGV) {
   my $filename = lc join '', (split '/', $url)[-2, -1];
   print qq{url: $url -> filename: $filename};
   }
"
/pdf/100004/_20090326.pdf
/pdf/100006/_20090326.pdf
/foo/bar/pdf/123/_456.PdF
output:
url: /pdf/100004/_20090326.pdf -> filename: 100004_20090326.pdf
url: /pdf/100006/_20090326.pdf -> filename: 100006_20090326.pdf
url: /foo/bar/pdf/123/_456.PdF -> filename: 123_456.pdf
[download]

Updates:

1. Added lowercasing ("\L") to  s/// version replacement string.
2. Added  split version.